Understanding Simplicity Bias towards Compositional Mappings via Learning Dynamics

Open in new window